Verda Cloud GPU Cost 2026: H100 and B200 Pricing vs Spheron

Verda's own pricing page listed H100 SXM5 at $3.81/hr on-demand on 7 October 2026. By 3 November 2026, the same GPU, on the same page, cost $4.841/hr. That's a 27% jump in under a month, and it landed right after the company closed a $117 million raise aimed at expanding past its Nordic home base. If you're pricing out Verda for an H100 or B200 job this quarter, the number you remember from last month is already wrong.

TL;DR: Verda Cloud GPU Cost in 2026

Verda's on-demand GPU rates have moved up sharply since its DataCrunch rebrand, and Spheron's live marketplace pricing is the no-contract comparison point for 2026 buyers.

Check current GPU pricing for live rates on both.

The DataCrunch-to-Verda Rebrand: What Changed and What Didn't

DataCrunch didn't get acquired, pivot product lines, or quietly fold into a bigger company. It renamed itself. According to Verda's own rebrand announcement, the new name combines the Spanish word for "true" and the Esperanto word for "green," tied to a stated goal of building a European hyperscaler focused on data protection, AI sovereignty, and 100% renewable energy. The infrastructure, the engineering team, and the Finland and Iceland data centers that made DataCrunch a credible EU GPU option all carried over.

What changed is the ambition and the money behind it. Verda (still operating as DataCrunch at the time of the raise) closed $117 million in funding specifically earmarked to expand the platform from the Nordics into the UK, US, and Asia. That's a materially different company than the one that built its reputation as a predictable, EU-only GPU shop for research teams and startups. A company spending to go global is also a company with a new reason to re-price its hardware, and the rate card backs that up.

For the fuller alternatives shortlist built before this rebrand, including the EU residency and billing-granularity tradeoffs that still apply, see 10 Best DataCrunch (Verda) Alternatives in 2026. That post's own price snapshot, taken 7 October 2026, is the baseline this post updates against.

Verda H100, H200, B200, and B300 Pricing: On-Demand, Spot, and Reserved Rates

Here's every current rate from Verda's own pricing page, checked 3 November 2026:

Spot runs almost exactly half the on-demand rate across every tier on this table, which is a cleaner split than most providers publish. On top of that rate card, Verda's pricing page lists reserved discounts at 1, 3, 6, 12, and 24-month terms. The exact percentage at each term isn't broken out on the public page; a team that wants the full schedule has to check current terms directly rather than budget off a remembered number, which matters more now that the base rate itself has moved twice in a month.

That reserved structure is the one lever Verda offers for bringing cost down without touching spot. There's no published multi-GPU node discount, no volume tier below the term-length discount, and no separate self-serve path to a lower number. You're either paying the on-demand rate, taking the spot risk, or signing a term.

Pricing fluctuates based on GPU availability. Spheron rates above are live as of 05 Nov 2026; Verda's rates reflect its most recently published rate card and may have changed. Check current GPU pricing → for live rates.

Why Finland and Iceland Data Centers Shape Verda Cloud GPU Cost

Verda's facilities sit in Helsinki and Reykjanesbær, chosen for a reason that shows up directly in the rate card: cold climate and renewable grid access cut the two biggest variable costs in running a GPU data center, power and cooling. Iceland's geothermal and hydro grid and Finland's cold ambient air both reduce the mechanical cooling load a facility needs to keep racks of H100s and B200s from throttling, which is the economic argument every Nordic-climate operator makes for running GPUs there instead of in a warmer region.

Verda doesn't publish a PUE (power usage effectiveness) figure on its pricing page or in its public documentation. The number worth knowing anyway is 1.0, the theoretical limit where every watt drawn goes to compute with none lost to cooling or power conversion; cold-climate sites like Helsinki and Reykjanesbær are built to get close to that floor without the mechanical cooling overhead a warmer region needs. Verda has said that its H200 deployment in Finland runs on 100% renewable energy with zero water usage, and the waste heat it generates feeds a local district heating system.

That's the upside case for why a European-only operator can compete on price at all against hyperscalers with far larger capital bases. It's also the constraint behind the November price moves: a single operator running its own facilities carries the capital cost of that infrastructure directly into its rate card, and when funding priorities shift toward geographic expansion, as they explicitly did with the $117 million raise, the GPUs already in Finland and Iceland are what gets re-priced to help pay for it. A marketplace aggregating capacity across multiple data center partners, by contrast, isn't carrying any single facility's capital structure into its number.

Verda vs Spheron: Same-Tier Cost Comparison

Spheron's pricing page describes a model built around the opposite tradeoff from Verda's: rates come from aggregating capacity across multiple data center partners rather than from one operator's fixed list price, billed per minute with a 20-minute minimum runtime, and the page states explicitly that there are "no commitments" and "no lock-in." That structure is why Spheron doesn't publish a reserved-discount schedule the way Verda does. There's nothing to reserve against; the rate itself is already live and marketplace-driven, so the discount Verda gets from signing a 24-month term is closer to what Spheron's spot tier and live on-demand pricing are already doing without the term.

The honest tradeoff runs the other way too. Verda's Helsinki and Reykjanesbær facilities are its own infrastructure, which means it can make a residency commitment Spheron can't: data that stays in a specific Verda facility stays there because Verda operates it directly. Spheron doesn't guarantee EU-only data residency the same way; for a team with a strict EU-member-state residency requirement, Verda, Nebius, or OVHcloud are the providers built to make that promise contractually, a distinction covered in more depth in GPU Cloud Providers in Europe 2026. Similarly, Verda's 2-25% term discount is a real, published lever for a team that can forecast 12 or 24 months of usage; Spheron has no equivalent, so a team that specifically wants a multi-year contracted discount rate won't find one on Spheron's pricing page.

Which Pricing Model Fits Your Workload

Three questions decide which side of this comparison actually matters for a given team.

Do you know your usage 12 to 24 months out? If yes, Verda's reserved tiers are worth pricing against a term sheet; a 25% discount on $4.841/hr brings H100 down to roughly $3.63/hr, which is a real number to compare against Spheron's live on-demand rate before signing anything. If your usage is spiky, experimental, or genuinely unpredictable, a term commitment is the wrong shape regardless of the discount, and Spheron's no-commitment on-demand and spot tiers fit better.

Does the job tolerate interruption? Both providers split pricing close to 2x between on-demand and spot. For batch inference, checkpoint-tolerant training, or anything that can resume from a saved state, spot is worth checking on either platform: Verda spot sits at $2.421/hr for H100, Spheron spot at $2.20/hr, against Spheron on-demand at $2.64/hr. Compare both tiers against the job's tolerance for interruption rather than assuming one is always cheaper.

Is B200 or B300 actually the right tier for the job? Sticker price alone understates what's different between a B200 and a B300 node. Running GLM-5.3 on SGLang with NVFP4 quantization at 256K context, an 8xB200 node configured as one 8-GPU copy served 324 tokens/sec for a single user and 117 tokens/sec at 32 concurrent sessions, with 0.90-second time-to-first-token. An 8xB300 node split into two 4-GPU copies, served fewer tokens/sec per user at every concurrency level, 254 down to 86, but handled twice the concurrent sessions, 64 versus 32, and halved time-to-first-token to 0.48 seconds. On Verda's own rate card, B300 runs roughly 28% above B200 per hour, and that premium buys session capacity and latency, not raw per-user throughput. A job bottlenecked on how many agents can hit the node at once is a different calculation from one bottlenecked on how fast a single stream completes, and sticker price doesn't tell you which one you're buying.

If reserved pricing, EU-operator-owned infrastructure, and a specific residency commitment matter more than flexibility, Verda's rate card is worth running the full term-discount math against. If the priority is a live rate with no term to plan around, Spheron's rate is the number to check same-day.

For a broader view of where these numbers sit against the rest of the market, GPU Cloud Pricing Comparison 2026 puts H100, H200, and B200 rates from Spheron, AWS, and Azure side by side, and Lambda Labs Pricing 2026 covers another provider running the same mix of on-demand and term-based discounts Verda does.

European sovereignty and renewable power are real reasons to pick Verda. If what you need instead is a live rate with no term commitment, Spheron's marketplace pricing updates continuously across multiple data center partners.

Check current GPU pricing →